AZ-305 ยท Compute solution design
Designing AKS clusters
AKS cluster design: node pools, cluster and pod autoscaling, network plugins, Microsoft Entra integration and workload identity, Azure Container Registry.
Key points
- System pool: critical system pods; user pools: apps (GPU, Windows, Spot).
- Cluster autoscaler: nodes; Horizontal Pod Autoscaler: pod replicas; KEDA: event sources.
- Azure CNI Overlay: pod IPs from a separate CIDR, saves VNet space, recommended.
- Azure CNI Pod Subnet: pods directly reachable; kubenet legacy, retires March 31, 2028.
- Workload identity: pods access Azure without secrets; ACR attach gives kubelet identity AcrPull.
Source: https://learn.microsoft.com/en-us/azure/aks/concepts-network-cni-overview
Practice questions
When you attach an Azure Container Registry to an AKS cluster with --attach-acr, which role is assigned and to which identity?
- AcrPush to the cluster control plane identity
- AcrPull to the kubelet managed identity
- Owner to the AKS resource group identity
- Reader to every pod's service account
Show answer
Answer: B. AcrPull to the kubelet managed identity
The integration grants the AcrPull role to the kubelet managed identity of the agent pool, which is what pulls images. Push rights are not needed to run images, and broad roles like Owner would break least privilege.
Which statements correctly describe AKS scaling components?
- The cluster autoscaler adds nodes when pods cannot be scheduled
- The Horizontal Pod Autoscaler changes the number of pod replicas
- The Horizontal Pod Autoscaler adds VMs to the node pool
- The cluster autoscaler changes the VM size of existing nodes
- KEDA can scale pods on event sources such as queue length
Show answer
Answer: A. The cluster autoscaler adds nodes when pods cannot be scheduled, B. The Horizontal Pod Autoscaler changes the number of pod replicas, E. KEDA can scale pods on event sources such as queue length
The cluster autoscaler changes the node count when pods cannot be scheduled or nodes sit idle. The Horizontal Pod Autoscaler changes pod replicas based on metrics, and KEDA scales workloads on event sources such as queue length. Neither autoscaler creates VMs from pod metrics directly or resizes existing nodes.
With Azure CNI Overlay, servers on-premises can open connections directly to individual pod IP addresses over VPN or ExpressRoute.
Show answer
Answer: False
False. In an overlay network, connectivity to on-premises is pod initiated only; traffic leaving the cluster is SNAT'd to the node IP. If on-premises systems must reach pods directly, use Azure CNI Pod Subnet or expose the pods through a service.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
